Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Audit and refine an already implemented interface for hierarchy, consistency, responsiveness, interaction feedback, accessibility, and visual defects. Use only in the review phase after the page runs; do not use to replace product requirements or initial design direction.
.claude/skills/qierkang-impeccable-ui-polish/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | -50% | 0% |
| case-22 | ✗→✓ | ▲ Improved | -58% | 0% |
| case-04 | ✓→✗ | ▼ Worse | -52% | 0% |
| case-11 | ✓→✗ | ▼ Worse | -75% | 0% |
| case-14 | ✓→✗ | ▼ Worse | -67% | 0% |
This is a safe project-local adaptation with no provider hooks.
Review product task and hierarchy; layout, type, color, and component consistency; all feedback states; responsive overflow; touch, keyboard, and focus; motion purpose and reduced motion; and evidence-backed performance risks.
Output findings first with file and behavior evidence. Fix product defects before preferences, rerun the affected build and browser workflow, then run scripts/util-verify-frontend-experience.sh when evidence is ready.
Do not install upstream hooks, run provider installers, or create or replace PRODUCT.md, DESIGN.md, AGENTS.md, or CLAUDE.md.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 4,283 | 18,712 | +337% | 1 | 1 | 0% | 155 | 291 | +88% | 0 | 0 | — |
case-07 | fail→fail | 14,814 | 4,834 | -67% | 1 | 1 | 0% | 3,241 | 517 | -84% | 0 | 0 | — |
case-02 | fail→fail | 42,243 | 4,869 | -88% | 1 | 1 | 0% | 8,235 | 344 | -96% | 0 | 0 | — |
case-03 | fail→fail | 4,753 | 3,744 | -21% | 1 | 1 | 0% | 295 | 275 | -7% | 0 | 0 | — |
case-04 | pass→fail | 6,887 | 6,092 | -12% | 1 | 1 | 0% | 1,140 | 551 | -52% | 0 | 0 | — |
case-05 | fail→pass | 18,808 | 7,634 | -59% | 1 | 1 | 0% | 3,136 | 1,570 | -50% | 0 | 0 | — |
case-06 | fail→fail | 2,658 | 6,752 | +154% | 1 | 1 | 0% | 117 | 685 | +485% | 0 | 0 | — |
case-08 | fail→fail | 7,511 | 5,383 | -28% | 1 | 1 | 0% | 1,365 | 500 | -63% | 0 | 0 | — |
case-09 | fail→fail | 7,044 | 4,387 | -38% | 1 | 1 | 0% | 1,140 | 430 | -62% | 0 | 0 | — |
case-10 | fail→fail | 12,767 | 4,680 | -63% | 1 | 1 | 0% | 2,159 | 476 | -78% | 0 | 0 | — |
case-11 | pass→fail | 13,406 | 6,302 | -53% | 1 | 1 | 0% | 2,049 | 506 | -75% | 0 | 0 | — |
case-12 | fail→fail | 12,643 | 5,233 | -59% | 1 | 1 | 0% | 2,169 | 467 | -78% | 0 | 0 | — |
case-13 | fail→fail | 14,827 | 4,391 | -70% | 1 | 1 | 0% | 2,409 | 293 | -88% | 0 | 0 | — |
case-14 | pass→fail | 9,439 | 5,490 | -42% | 1 | 1 | 0% | 1,362 | 448 | -67% | 0 | 0 | — |
case-15 | fail→fail | 14,196 | 5,062 | -64% | 1 | 1 | 0% | 2,267 | 431 | -81% | 0 | 0 | — |
case-16 | fail→fail | 13,552 | 5,972 | -56% | 1 | 1 | 0% | 2,260 | 469 | -79% | 0 | 0 | — |
case-17 | pass→fail | 14,049 | 5,337 | -62% | 1 | 1 | 0% | 2,294 | 423 | -82% | 0 | 0 | — |
case-18 | fail→fail | 10,235 | 4,915 | -52% | 1 | 1 | 0% | 1,566 | 406 | -74% | 0 | 0 | — |
case-19 | pass→pass | 9,952 | 4,836 | -51% | 1 | 1 | 0% | 1,573 | 736 | -53% | 0 | 0 | — |
case-20 | fail→fail | 13,153 | 3,924 | -70% | 1 | 1 | 0% | 2,368 | 318 | -87% | 0 | 0 | — |
case-21 | fail→fail | 16,161 | 5,203 | -68% | 1 | 1 | 0% | 2,741 | 439 | -84% | 0 | 0 | — |
case-22 | fail→pass | 10,846 | 3,342 | -69% | 1 | 1 | 0% | 1,562 | 650 | -58% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 3 counted toward the lift figure. The other 19 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of -9 percentage points is the difference between those two pass rates over the 3 comparable cases. 5 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.